FPGA does not have "performance" like a processor. This is just a bunch of logical elements (LEs) that you can use. If the high-end part has 2MLE and the younger part has 200kLE, but you only need 20kLE for your processing core, it does not very much depend on which one you use, all other things being equal. Of course, if you have a problem that can be easily parallelized, you can turn these additional LEs into extra performance by building more processor cores. But you have to do it.
Now everything else is not always the same, because FPGAs are much more than just the number of logical cells. I can’t talk about Xilinx components (I work for another major FPGA vendor), but usually upscale families will have things like very high speed transceivers that aren’t in the middle and younger families. In addition, sometimes they have different combinations of internal memory, DSP, etc.
So can you use a bunch of small FPGAs instead of a large one? Remember that an FPGA will only have about 1000-2000 IO, while there will be more than 100K internal wires between the corresponding parts of the top. Thus, you will not only need to create a rather complex board, you may find yourself IO-limited in receiving signals from one chip to another.