Lattice conformation performance
The input data conformation process in v2.x is very thorough and can correct most geometry errors, but has mediocre performance characteristics, particularly for large input data sets.
A first band aid for this problem would be the inclusion of multiple conformation complexity settings, including one where this step is disabled completely. This leaves room for input data sets of any size (assuming the data is sufficiently clean for Triangle).
Long-term, the conformation step should be revised to scale more favourably with large data sets, likely through introduction of a spatial index or other mechanisms to reduce the data set size locally.