This article and the benchmarks are more applicable to PS5 since it's going to be mostly Kraken + Oodle Texture....
https://cbloomrants.blogspot.com/2020/07/performance-of-various-compressors-on.html
Interesting it's using 256K chunks as a supposed "typical" game usage. So maybe PS5 will also do 256K blocks which would be nicer on the FTL requirements, but OTOH it would require more expensive SRAM buffers on the hardware decompressor. It's hard to tell.
Using large chunks improve the compression ratio. If they go for 32KB it would improve latency and IOPS, but compression would logically be worse. I am convinced they entire hardware pipeline would need to be designed with a fixed block size in mind, but we don't know what. I think it could be anywhere between 32K and 1MB.
There are still a lot of question surrounding it, and I would hope DF will address it at some point. Maybe none of their contact want to say anything, or they can't corroborate this, or DF signed an NDA, who knows.
The typical data of games will vary wildly, so will compression. But they can crank the lambda way above 40, it seems that lambda 40 is "mostly" imperceptible from lossless, some types of data like things with text might want it more lossless. Also we have benchmarks showing geometry will compress 3:1 on average with Kraken.