According to Wiktionary, the words for 3 and 4 in Proto-Slavic are *trьmi and *četyrьmi, respectively, in the instrumental case. In (current) Czech, they evolved into třemi and čtyřmi. But if you applied Havlík's law (or the naive version thereof, anyway), you might guess that the instrumental of the Czech word for 3 is *třmi. Arguably, we have třemi as an exception to the "odd yer" rule so an awkward consonant cluster "třm." But it is not at all clear to me (1) exactly what consonant clusters the Czech language considers as awkward and (2) how it avoids them. Is there a reference to a general rule that can explain the aformentioned morphological difference?
The form "třmi" is actually documented as the Old Czech form of the number
třmi = třemi
Zdroj: Šimek, F., Slovníček staré češtiny. Praha: Orbis, 1947.
Dalimil: "Na léto Tateři pojidú a třmi prameny vnidú."
Gebauer in his Historical Czech grammar, Part I, point 66 "trt, tr̥t from trьt" shows trьmi -> třmi as an example of this regular change with the exception that we get ř instead of r, probably by analogy to other cases (tři, třech,...).
I am not sure "třm" is that awkward on its own that it must be avoided considering the still current noun "třmen" (also as "střmen" existing as a toponym "Střmen") and the diminutive "třmínek". Even if it was finally changed to třemi, the form "třmi" was surviving for a long time well into the 15th century and can even be encountered in the Kralická bible.
As to possible consonant clusters in Czech in general you can refer to Czechency: Fonotaktika (in Czech) or Bičan, Phonotactics of Czech (PhD thesis) (in English). The latter says "That the voiceless consonants can occur before /řM/ is proven by /třMen/.".