I've read through various documentation on consensus algorithms, along with the fellow SE post What is the difference between Babe, Aura, and Grandpa?. I'm unsure what the posited advantage of BABE/GRANDPA is when you have equal weights.
The only possible idea I may have is GRANDPA's chain finalization, instead of block finalization, offers greater efficiency despite use of VRFs and the amount of forks it causes, along with a consistent block time (and reduced signature count). With 50 validators at 5 seconds, a 30% failure (below the expected BFT threshold) would cause a 75s latency at worst which could be considered unacceptable. Having 30% of nodes fail at once (and right next to each other in the queue) should almost never happen though when discussing real world usage? Perfectly even distribution would have every second block take 10 seconds (5, 5, failure, 10, 5, failure, 10, 5, failure, 10, 5...).
Are there any other considerations I'm missing? Has there been a performance analysis publicized comparing the two? Are the consensus algorithms sufficiently light weight where performance isn't considered a concern and the added robustness of BABE/GRANDPA makes it optimal?