Win1251 charset contains some Greek characters that are not explicitly paired with the upper/lower counterparts. For example, the small letter "mu" (code 0xB5) doesn't have a capital counterpart. I don't really know how to explain that -- whether this is as designed ("mu" actually means a scientific symbol) or it's supposed that the Cyrillic capital letter "em" (code 0x4D) should also act as a Greek capital letter (their visual glyphs are the same).
Anyway, the small "mu" is successfully uppercased by the v2.1 Unicode internals, but its capital counterpart doesn't fit Win1251 when applying the characters backward. In other words, upper(_win1251 '<mu>') produces transliteration error in v2.1. In practice, it means that some scientific strings (where "mu" usually means "micro") cannot be uppercased or searched with CONTAINING in v2.1.
In prior versions, "mu" was never uppercased. I would suggest to keep this behavior.