I could not find any information what is the relation between the AmpEG Release numeric values and actual time in sec/ms. Can anyone enlighten me?

That information is not published in any of the documentation. The AEG (Amplitude Envelope Generator) consists of the following components:
Initial Level
Attack Time
Attack Level
Decay 1 Time
Decay 1 Level
Decay 2 Time
Decay 2 Level
Release Time
These are the parameters that the synth uses to describe what in old analog synthesizers was referred to as the ADSR (Attack-Decay-Sustain-Release). It uses a series of TIME and LEVEL settings to describe the loudness shape of the sound.

In general, the Amplitude EG, Level is the output, minimum to maximum, 0-127. Time is how long it takes to reach the next Level setting, where 0 is immediate, and 127 is a longer time. The Time parameter is not linear. Designed specifically for musical instrument emulations, the time difference is weighted so that its density is greatest in lower values and extremely sparse as you go higher. The Time difference between 56 and 57 is way, way shorter than the time distance between 125 and 126.

AEG is set by ear, not by seconds and milliseconds.

I’m not exactly sure what you are “converting some SFZ” (is that a file format?) or are you trying to recreate a sforzando articulation (?) Sorry this is not clear from your post.

Let us know... MODX offers some unique methods to emulate the musical gesture Sforzando (none of which requires a stop watch or a chart!)
You can automate it, you can setup for manual control, you can work in realistic timbre change along with the amplitude change.
The brass swells and sforzando gesture emulation on this engine break some new ground...imho

Hey, Bad Mister, thanks for the reply.

Yes, SFZ is a file format, it's extended version of the good old SF2 (Sound Font) that was pretty popular in the 90s... It basically consist of samples and a script for the player/sound engine. The SFZ format is using explicitly real life metrics for all parameters - seconds, dB, Hz, percentages, so to replicate the original programming on the MODX I need to know approximately what Release values are close to 0.5s, 1s, 5sec...
Is that a proprietary and confidential information?

2017: https://www.yamahasynth.com/ask-a-question/release-time-duration

... led to experimental brute-force experimentation

2018: https://www.yamahasynth.com/ask-a-question/time-table/oldest

If this is proprietary or not can be debated - but it seems to be a topic that will not receive detail from official sources.

Thanks Jason, that's exactly what I was looking for!

On a side note, speaking of "musicality" and setting EG by ear, when there are so many values that produce essentially the same audible result, is sort of cynical. I can picture in my head the poor sound designer that spends hours trying to decide if attack value 2 or 10 sounds better on this snare hit...
Seriously, before you can trust your gear and set parameters "by ear", you need to know your gear!
With all the things Yamaha is doing great, there are still some bits that need to be pulled off the 80s and set to meet the third decade of the 21st century.

There's plenty of musical reasons to approach any given problem from more than one angle - numbers, math, graphs, and non-ear ways included. And there are all kinds of reasons why one would try to convince the curious musician otherwise.

Yamaha has collaborators that use their ear as only one tool in their arsenal and use math, numbers, graphs, non-audible methods as well. There's room for both and the intellectual approach is propped up and celebrated under different circumstances. "Know your audience".

I'm not sure philosophy moves the needle (doesn't offer any traction to progress the answers). If you can't get your information from the source - you'll have to look elsewhere or construct ways to find the answers from the equipment.