plx benchmark  (PostgreSQL 18.4, min of 5 runs)

workload sizes: arith=2000000  strbuild=200000  iter=1000000 rows  branch=2000000  call=500000 (perl/py call=50000)

language      arith           strbuild        iter            branch          call          
------------------------------------------------------------------------------------------
plpgsql       48 ms (1.00x)   1597 ms (1.00x)  88 ms (1.00x)   150 ms (1.00x)  145 ms (1.00x)
plxruby       50 ms (1.06x)   9 ms (0.01x)    97 ms (1.10x)   153 ms (1.02x)  151 ms (1.05x)
plxphp        52 ms (1.09x)   9 ms (0.01x)    99 ms (1.11x)   154 ms (1.02x)  149 ms (1.03x)
plxjs         51 ms (1.07x)   10 ms (0.01x)   94 ms (1.06x)   154 ms (1.03x)  153 ms (1.06x)
plxpython3    53 ms (1.12x)   10 ms (0.01x)   99 ms (1.12x)   154 ms (1.03x)  149 ms (1.03x)
plperl        42 ms (0.88x)   12 ms (0.01x)   507 ms (5.74x)  139 ms (0.93x)  273 ms (1.89x)
plpython3u    64 ms (1.35x)   14 ms (0.01x)   337 ms (3.81x)  108 ms (0.72x)  165 ms (1.14x)

x = relative to plpgsql (lower is faster). plx dialects transpile to plpgsql.
Note: perl/python 'call' uses 1/10 the iterations (per-call SPI is far slower).