My general rule is to reuse a variable when the old value is no longer needed and the variable name also describes the new value, so I would prefer 2A but the comment explaining that $module is to be a canonicalized module name is very important. Creating additional lexicals is cheap, but not free in Perl. (Additional locals are essentially free in most cases in C since modern compilers allocate the entire stack frame at once.)

My fellow monk choroba made a good point about filtering input when you can, but I would also prefer 1A because that type of filtering at the beginning of a loop's block is idiomatic in Perl. Concision in this case is also useful in that the more concise code requires fewer VM steps because it avoids an extra lexical. Filtering the input is the best option, since grep iterates in C and reduces the number of iterations perl's VM must execute. This is a trivial concern in most cases, but can be serious in an inner loop.

Lastly, I think you meant "next unless -d $dir" in 1A, 1B, and 1C — "next if -d $dir" skips the iteration if $dir does name a directory and would be very confusing in all three cases.

Edited by jcb: Add missing caveat; thanks to GrandFather for pointing out my mistake.