It appears that there are still some differences when you run make reftests on the same commit but different systems. This is almost certainly related to randomness somewhere. It appears to exist in these different areas:

• ssz_generic/progressive_containers/invalid/*modded*
• sanity/blocks/pyspec_tests/mix_blob_tx_and_non_blob_tx
• operations/proposer_slashing/pyspec_tests/builder_payment_deletion_current_epoch
• operations/proposer_slashing/pyspec_tests/builder_payment_deletion_previous_epoch
• operations/proposer_slashing/pyspec_tests/builder_payment_deletion_too_late