The sorting phase of compression gathers together similar strings in the file. Because of this, files containing very long runs of repeated symbols, like "aabaabaabaab ..." (repeated several hundred times) may compress more slowly than normal. Versions 0.9.5 and above fare much better than previous versions in this respect. The ratio between worst-case and average-case compression time is in the region of 10:1. For previous versions, this figure was more like 100:1. You can use the -vvvv option to monitor progress in great detail, if you want.

Decompression speed is unaffected by these phenomena.

bzip2 usually allocates several megabytes of memory to operate in, and then charges all over it in a fairly random fashion. This means that performance, both for compressing and decompressing, is largely determined by the speed at which your machine can service cache misses. Because of this, small changes to the code to reduce the miss rate have been observed to give disproportionately large performance improvements. I imagine bzip2 will perform best on machines with very large caches.