Performing a g-nmg conversion of a moderately "complex" model
such as havoc.g with any normal tolerance listed (tested 1, .01,
15, 0.1) results in the application apparently either getting stuck
glueing faces together in an infinite loop, or there are simply too
many faces being generated/tested.
The O(N^2) performance is troublesome but "shouldn't" be the
issue especially with a normal tolerance of 15 (assuming -N means
degrees like the manpage states). It's worth noting that simply
removing the normal tolerance makes the conversion complete as
expected (albeit after about an hour).