I don't an answer to your question, but an obvious difference is that
the "slow" query contains many more loops. (this may already have been
noted, I didn't see it posted however).
(showing just the loops with more than one loop)
-> Index Scan using assemblies_pkey on assemblies a (cost=0.00..0.31
rows=1 width=38) (actual time=0.007..0.009 rows=1 loops=3685)
-> Function Scan on stockperowner_lead_ab c (cost=0.00..15.00 rows=5
width=20) (actual time=0.012..3.162 rows=1694 loops=3685)
-> Index Scan using idx_u_assidpartid on partsassembly b
(cost=0.00..0.27 rows=1 width=16) (actual time=0.010..0.012 rows=1
loops=3685)
-> Hash Join (cost=97.69..531.29 rows=1250 width=21) (actual
time=2.395..78.855 rows=3851 loops=3684)
-> Seq Scan on allocatedassemblies b (cost=0.00..349.62 rows=19062
width=12) (actual time=0.009..35.493 rows=19062 loops=3684)
-> Hash Join (cost=621.63..1206.10 rows=3854 width=28) (actual
time=0.074..71.265 rows=593 loops=3684)
-> Hash Join (cost=337.11..839.61 rows=3860 width=32) (actual
time=0.057..68.467 rows=593 loops=3684)
-> Seq Scan on poparts e (cost=0.00..379.60 rows=16860 width=32)
(actual time=0.008..34.510 rows=16860 loops=3684)