I agree with your choice and mostly agree with your comments.
The one disagreement is that the opaque server connection doesn't require creating homework assignments in WeBWorK (as LTI and wwassignment do). With opaque the homework/quiz assignments are created using the moodle quiz object -- with each individual question being supplied from the webwork OPL or optionally from a collection of .pg files created locally on the opaque server or stored in the moodle question object. The moodle quiz module provides more, or at least different, flexibility than either the webwork homework assignment or the webwork gateway quiz.
However I agree with the choice of using LTI because in my opinion the opaque server is just barely ready for prime time. It was commissioned by the late Mika Seppala for WEPS (world educational portals)
(now at https://geom.mathstat.helsinki.fi/moodle/) and has received only modest testing. The purpose was to allow STACK and WeBWorK to be used interchangeably as backend engines for evaluating math questions. It works but could still use cosmetic interface improvements.
On the plus side the technology in the opaque server is essentially the same as that which provides rendering for the MathBook_XML (PreTexT) project (http://webwork.maa.org/moodle/mod/forum/discuss.php?d=4194) and was the base on which that engine was built.
My only reservation about LTI vs. wwassignment is the current limitations on the grade data passed back by the interface (just one number per url or "tool").
wwassignment is slightly more flexible. Future versions of LTI should correct this at which point I expect that the rationale for wwassignment will go away.
I run a moodle installation here at the U. of Rochester and use it myself, but it is not widely used by others at the university so I don't get a lot of feedback with respect to the moodle/webwork interoperation, nor am I working with people with extensive moodle expertise. I'd be happy to work with any IT or math people running a production moodle installation about improvements that can/should be made to wwassignment or to the opaque connection.
-- Mike Gage gage@math.rochester.edu