With the NumericUnderscores language extension, 100_000_000 is valid Haskell syntax. Without the extension, it is not, so lhs2tex doesn't need to be aware if the extension is turned on or not.

The problem is that lhs2tex tokenizes 100_000_000 into two tokens: a numeral "100" and a varid "_000_000". Even if varids and numerals are otherwise set up to be typeset the same, it is a problem because it adds an extra space between "100" and "_000_000" in the output.