FWIW, I did test your exact track lengths which I added to a fake cuesheet and used foobar2000 + the foo_musicbrainz component which calculates discids.
Actually this is not pretty common. The discid is basically SHA-1 checksum of the TOC. SHA-1 has 160 bits, So given two different arbitrary TOCs the probability that they both generate the same checksum is 1 / 2^160 (~10^-48). And even if you take the probability of a collision in a larger data set, let’s say 1 million different TOCs, the probability is still less then 10^-36 if my math was correct. MB currently has ~825k disc IDs, it’s rather unlikely that there is a collision with different TOCs.
What is actually much more likely is two completely different discs having the same TOC, and there had been a few cases of this as far as I remember.
And of course different releases of the same album often have the same disc ID. But this has not much to do with probabilities, but more with production processes
It appears EAC uses the format mm:ss.MM
[…]
(where MM is milliseconds)
MM here in your EAC logs is not in milliseconds. The unit is “sector length” (1/75 of a second). This is typical for applications displaying timestamps from CDs.
There is no need to decode these length fields to calculate disc IDs as they are merely human-readable presentation of the information from the start/end sector columns.