I suspect that what he notes as “non-linear scaling” (or whatever term he used) comes mostly from Hugo’s current “eager loading everything in memory” approach, which works great in common use cases, butt slows down/fails in the really big sites.
I have a patch nearly done (that will be followed up with another one related soon after) that I think will dramatically improve this situation – as it introduces a more streaming approach, dispatching data in and out of memory when needed.
It would have also been interesting to see how more advanced features (bundling capability, image processing, module-like loading) would change things/ further separate.
gotta admit, I knew Hugo was the fastest, I just didn’t realise it is fastest by that much!
Looking forward to see the feature “Pages from Data” 