Largely because a lot of compression is contextual to the data around it. That makes most decompression a poor fit for parallelism.

The primary choices are space savings and power efficiency. Dedicated hardware and serial decoding/encoding often win out as a result.