Forgot to mention, *ds* varies from 0 to 1. It is the fractional part of the scale exponents, so when it reaches 1 we have the same set of scales and a perfect loop. If there was an infinite number of layers, we could change *ds* from -inf to +inf.
The machine was a 16-core/32-thread Linux box with 384 GB of RAM, running Kim Wallisch's parallel implementation of the Deleglise-Rivat algorithm (mentioned, but not benchmarked in the linked paper).
Hi Henrik & Daniel! Thank you both very much for your time and for the valuable contributions to my question, i really appreciate that! I wish you both also all the best! I am sure i will need these solutions once upon a time as well, and many others...