Paul - I found a very similar issue here:
community.sdl.com/.../9717

I'll highlight this again - the Avira antivirus which I run is not the issue, I've even tested the performance with the PC isolated from all networks and with the AV completely turned off; the problem persists, all the same for the AV will all the exceptions added as instructed in the KBs you've posted. The Translation Results window shows "Searching..." for over 3 seconds, usually up to 10 or 20, even with two or three-word segments. I didn't see anything like this in Studio 2014.
Also, the TM is new from scratch and being populated on the go. I've run fuzzy index recompute, to no avail.

I also noted that when I hold the cursor on a previous segment, and then skip to the next empty segment while the Translation Results shows "Searching...", the search slowly returns a fuzzy match for THAT PREVIOUS SEGMENT, not for the next empty one (which partially matches a completely different source, saved to the project TM, say, 10 segments ago). That is absurd.

This is unacceptable at work on jobs with thousands of words in partials (I am already behind schedule with a large project and WILL have to pull an all-nighter :( ). I've never seen something like that in the worst-performance CAT tools, like Across with a TM on a client's server and laggy pings on its end.

Paul 
1. No. All TBs and TMs are on a single logical system partition which occupies a single physical HDD.
2. No, see above.
3. What I did with the TM which revealed the issue was to disable upLIFT and reindex the TM. It changed nothing. What exactly would you suggest to turn off in the Bells and Whistles Department, to restore the neat performance of TM lookup and CM/70-100% match application that I did enjoy in Studio 2014?

Also, when the Translation Results is sluggislhy trying to keep the pace with me jumping the cursor between the segments in Editor, Studio often freezes and crashes to desktop without any error message.

Tip: the source document is an XLS spreadsheet with just one column and ten thousand or so rows.
I'm at row 5000-ish, which is about 4000 pairs in the TM now. The TM was made fresh for the job.
The context is that many segments are farming equipment part descriptions, so they contain numerical placeables and you can have five to twenty segments with a word or two of any difference in between them. Alas, the TM lookup mulls over a, say, 90-percent match for dozens of seconds before inserting a fuzzy.

Here's the fun part: I've added a different project with an MS Word file as a source. Three TMs are hooked up, and each has tens of thousands of segments. They're not even upgraded to Studio 2019, and the Translation Results runs like the Roadrunner trying to rile up Wile E. Coyote.