Can I get you in touch with one of our two developers working on this issue full time right now?
The issues we’re encountering remind me of the arbitrary 500 errors when accessing certain pages of products via REST from certain stores. If I remember correctly, there was some kind of indexing fault in that case but some stores are just hitting some kind of wall also. e.g. we have to restrict our page size to well below the advertised maximum which reduces efficiency in the eyes of the seller because they have to wait longer.
In one case, we see unpredictable 50+ minute delays for a simple handful of orders. This prevents us from running any product bulkQueries during that window. With 3-4 subsystems all depending on bulkQueries we can’t arbitrarily cancel any bulkQuery taking longer than 10 minutes or some processes may never complete.
In another case, we just get arbitrary timeouts which force us to retry bulk and regular queries with half the request size e.g. first = first / 2 then retry or for bulk, half of the chunk of IDs we tried last time (again well under the advertised 250 limit - think 10-20 at a time 
)
Specifically on the bulk side, only getting partial results or seeing frequent bulk query failures forcing robust retry logic.
On the plain graphql side, this would involve complex multi-tiered nested queries to ensure we retrieve 100% of the connections under a product/order which is high-risk for making mistakes.
Bulk would be preferable but the performance/reliability is just not there.