I have a client with a FM DB that might be considered large. 880 tables, 2300 table occurrences, 32 million records (a couple tables with over 12 million records each), 35,000+ lines of code, almost 4,000 layouts, 500 concurrent users.
I do NOT recommend moving to a MariaDB or it's ilk. FM is quite well optimized for performance, IF you make the right PRO CODE design decisions (some of the low-code ease of implementation factors do NOT scale well).
And it is [now] fast (once we redesigned screens and code). The original designer/coder's extensive use of unstored calculations, and Execute SQL, took a serious toll on performance. One ExecuteSQL line in one script, took 6 min 47 seconds to execute. We replaced it with a TOG with the appropriate relationships, and that line ran in .13 seconds. (note; ExecuteSQL will hang on open records in the data set being used, until that record is free - serious performance hit).
I would suspect, as others have alluded to, that there are layout objects on the navigated layouts that are negatively impacting performance - the DB in question is on the small size. Unstored calcs and summary fields are the primary culprits. There are work-arounds without losing the desired functionality. (the problem with low code, is it makes it easy, but not necessarily performant - that's were PRO-code comes into play).
@Malcolm hit on the key elements - unstored calcs (alt: replace with a auto-enter, LET triggered construct), sorts with un-indexed fields or based on normalized data from related tables (an occasion for denormalizing the data structure), summary fields (only show them on demand when needed, NOT as part of the default layout), going to a blank layout to perform context activity as no record content gets loaded, entering FIND mode before switching layouts, then switch to browse mode,, ideally on a subset of records (the initial layout load would otherwise load ALL records, THEN reduce it to the found set. Entering FIND first, removes this big transfer.
There is a ton of other more technical performance tweaks that can be used through judicious design practices.