CRC32 weirdly slow? (slower than both MD5 and SHA1)

[NintendoManiac64]

When re-investigating the performance of MD4 in the last version that supported it (v2.2.3), I found out that in both v2.2.3 and your latest v2.5.0, CRC32 is actually slower than both MD5 and SHA1 when tested on two completely different PCs.

　
All systems were configured to 2core/2thread with 4x DDR3-1333 in dual-channel reading the 48GB of data from an SK Hynix Gold S31 500GB connected to a motherboard supporting 6Gbps SATA, and the resulting checksum file was written to an uncompressed ramdisk.

[v2.5.0] Athlon II x2 250

115 sec - BLAKE3
116 sec - MD5
121 sec - SHA1
123 sec - CRC32　◀ ◀
166 sec - SHA3-256
187 sec - SHA256

[v2.5.0] Xeon E3-1246 v3 (basically a Xeon-badged i7-4790 non-K)

100 sec - BLAKE3／MD5／SHA1 (I/O limited)
107 sec - CRC32　◀ ◀
109 sec - SHA256
120 sec - SHA3-256

[v2.2.3] Athlon II x2 250

133 sec - MD4
149 sec - MD5
160 sec - CRC32　◀ ◀
181 sec - SHA1
464 sec - SHA256

[v2.2.3] Xeon E3-1246 v3 (basically a Xeon-badged i7-4790 non-K)

111 sec - MD4
125 sec - SHA1
127 sec - MD5
138 sec - CRC32　◀ ◀
325 sec - SHA256