A student and I actually did a
performance study on crypto accelerator cards. The long and the short of it is that these cards help somewhat, but don't expect a massive improvement. In a later version of the study (not yet published), we compared the performance of dual-processor systems with single-processor + accelerator and several other variants on that theme. In most cases, having a second CPU will increase your throughput more than having an accelerator card. Of course, having two CPUs
and an accelerator will be even better.
It's all about dollars. If it's cheaper for you to add an accelerator card, do it. If it's cheaper for you to add another CPU, do that instead.