Disclaimer: I make software which tends to run on mass-market processors. So I'm not familiar with the kind of niche markets where FPGA would make sense.
It's very hard to design a cryptographic algorithm. Designing something that works functionally is easy, but designing something secure is hard — and if you don't need something secure, there's no point in doing any cryptography. As a consequence, cryptography tends to use only a small number of standard primitives. This is further exacerbated by the fact that one of the biggest use cases of cryptography is communication, and this requires all parties to agree on the algorithms, thus again standardization around a small number of primitives.
As a result, there's no room for niche primitives. So if something is worth accelerating, it's typically worth either designing an ASIC or even expanding a general-purpose processor. For example, AES acceleration is available on most high-end processors (x86, arm, …) and a growing number of microcontrollers. This is increasingly the case for SHA2 as well. Acceleration of public-key cryptography is less widespread, but common on some markets such as smartphones and some microcontroller market segments.
I can only think of two niche applications where dedicated circuits are worthwhile. One is brute-force work, for applications such as password cracking and cryptocurrency mining, which typically uses standard primitives in a very specific way where it can be advantageous to make a dedicated circuit for that specific way. For these applications, all of off-the-shelf CPU, off-the-shelf GPU, ASIC and FPGA are in the running, depending on the nature and amount of work. Another application is experimenting with new algorithms or new implementation techniques. For that use case, FPGA are definitely worthwhile — then you design a specialized circuit (or not) based on observations made on the FPGA.