In the new rand, i'm seeing evidence of uneven bias at huge ranges, but not the same kind as before. The bias seems identical for both $rand and $rands, so it's more likely caused by identical post-processing of the random inputs.
That is very likely the reason for it. In both cases, mIRC needs to round an int64 to a double, as doubles are used in most calculations, eg. $calc().
On the other hand, are you saying that when $rand() is used with values in a reasonable range, ie. not huge values that are going to overflow due to internal conversions, that it is working well?
Regarding %throwaway_above, I am not convinced that this is something mIRC should be doing. The scripter can easily cause calcuations to overflow throughout the scripting language, if they choose large enough numbers.
Or are you saying that $rand() and $rands() should be mapping all int64 results to a specific maximum value?