"OCP behavior" as a presumptive fact was actually invented much later than Goldsmith's work, peaking around the time of the rise of OT. There is a difference between Goldsmith and Leben, centering around this OCP thing, which is remotely connected to what we now term OCP behavior, which I'll explicate.
Leben's theory (his 1973 MIT dissertation) sets forth the precursor of autosegmental theory, whereby in some languages there were two feature matrices, the segmental and the tonal: each was a standard SPE feature matrix, and the matrices were only indirectly connected, by being a fact of a given morpheme. (That is, vowels didn't have tone, tones were not on vowels). Complex rules of "mapping" would, in the derivation, assign the kth tonal feature to the ith segment. Tone melodies were essentially the same as word-class diacritics. The concept which morphed into OCP emerges in L's dissertation in 2.3 with his analysis of Mende noun tone, and the now infamous melody analysis whereby nouns (mostly) select one of the patterns H, L, HL, LH, LHL. He asserts that
because contours such as HHL and HLL have the same "melody", which is
represented suprasegmentally as HL, it follows that a language in
which tone is suprasegmental ln underlying forms will not be able to
have distinct contours HHL and HLL in underlying representations.
The problem is that this "is the same melody" claim is asserted and not proven. What this buys you in Mende is that it gives you a mechanism to account for the presumed combinatorial limits on tone (the problem nouns are somewhat exceptional, so he's not discarding masses of data, and the "exceptions" are generally set aside). This is the original seed of the OCP, but was not named by Leben and was not claimed to be a universal representational stipulation. (I believe that he saw it as a consequence of the kind of abstraction from surface forms that would lead to positing that a given language represents tone suprasegmentally). Overall, his analyses eliminated the need for underlying multiple instances of H or L within morphemes, since he found that they could all be derived by a rule which basically repeats the last tone in the tonal matrix. The crucial characteristic of Leben's system is that it has no underlying information on where a tone is linked, and he marks tone melody as a diacritic feature on whole morphemes.
Where Leben's theory invoked a "tone pattern diacritic" plus underspecification of tone features (filled in by copying rules, assigning tone into the segmental matrix by reference to the tonal matrix), Goldsmith posited full and equal partnership between tones and segments. It emerged that the most straightforward account of languages that he dealt with would necessarily generate "OCP violations", for example his derivation of Lomongo in sect 3 ch. 1 assigns two adjacent H tones to the first syllable of a H verb root plus H suffix. The OCP is explicitly identified in ch 1 sect 4.2 where G states "the inclusion of Leben's principle (which I shall call the Obligatory Contour principle) leads to unnecessary complications". The specific wording proposed by G is
At the melodic level of the grammar, any two adjacent tonemes must be
distinct. Thus HHL is not a possible melodic pattern; it automatically
simplifies to HL.
This is intended to represent Leben's beliefs, with the condition that for Leben this would be a fact about underlying forms but derived representations. While L does not explicitly say this, it is observationally consistent with what L does, and L has not to my knowledge objected to this as a mischaracterization of his views at the time.
G's argument against the OCP is a straightforward necessity argument: in order to account for certain of Tiv's inflectional tone patterns, it was necessary to posit HH as a morpheme-internal sequence. Underlying melodies with OCP violations are also necessitated for Etung, which contrasts the pattern LLH and LH, or HHL and HL, the problem being to say where the first tone is associated. G has available the mechanism of "accent" whereby a given vowel could be "accented" and this the initial locus of tone association. G is not satisfied with the conclusion that Etung is an "accent" language, thus OCP violation is taken to be necessary.
G posits that "structure" is more highly valued than "substance", so given a representational ambiguity between treating [bákí] as two H tones vs one H linked to two vowels, the latter representation is preferable (simpler), though language facts could prove that two Hs are necessary – both are possible representations, but OCP-obedience within a morpheme would be the acquisitional default.
So: (1) OCP, at the time, had very little explanatory force and was not clearly articulated as a strong principle of universal grammar until later (by McCarthy), and in tonology was there only to explain pattern restrictions – the reasons to assume the OCP were very weak to start with; (2) such pattern restrictions were found to not be universal even in the set of "suprasegmental" languages. The true rise of the OCP as a social phenomenon is McCarthy's analysis of Arabic root patterning, in his dissertation.