A few days ago I've changed sshx/scpx methods on MS Windows to use
powershell as local encoding shell. With this, the template
RemoteCommand='%l' seems to work well.
Would you mind to test it in your environment? Honestly, I'm kind of
restricted, because I don't run MS Windows on my machines.
It works! One thing I noticed while testing, however, is that there are a couple of cases where the latest git revision is quite a bit slower than Tramp 2.5.0.3. This might be due to the switch to Powershell, or perhaps something else has changed.
The biggest performance issue that I saw is exiting from M-x gdb[1]. Using sshx on Tramp 2.5.0.3 (with the quoting fix for RemoteCommand), after M-x gdb, I can hit C-d in the gud buffer to stop debugging, and it's almost instant. On the latest git revision, Emacs hangs for a couple seconds after hitting C-d.
Another particularly-noticeable case is disabling auto-revert-mode in a remote buffer (again using sshx). That too was almost instant on 2.5.0.3, but takes a couple seconds on the latest git. Enabling auto-revert-mode is fast on both versions, although I think 2.5.0.3 is still faster. Other things might also be slower on the latest git, but nothing else stuck out during my tests.
All the above cases were tested with a Windows 10 system using the default MS-provided OpenSSH client and connecting via sshx to a Linux VM running on localhost. I tried with scpx too, and things seem to be the same there. I'll send some Tramp debug logs off-list for the M-x gdb test in case they're useful for you (I don't think there's any sensitive information in them, but better safe than sorry).
Thanks,
[1] Caveat: there are a couple issues in core Emacs preventing M-x gdb from fully working over Tramp, but they shouldn't affect this test. I'll be submitting patches for these problems once I get them extracted from my .emacs and cleaned up a bit.