Theoretical minimum latency of H.264 vs MPEG-2

[Blue_MiSfit]

A buddy of mine working in a sports broadcast facility posed a very interesting question I had a hard time forming an opinion on:

GIVEN
H.264 4:2:2 intra-only (could be 8 or 10 bit)
vs
MPEG-2 4:2:2 intra-only

Encoding and decoding 1080i59.94 live from / to HD-SDI

THEN
Which encoding standard could deliver the lower theoretical end to end latency?

His shop almost exclusively uses Ateme FPGA based encoders and decoders via managed networks (dedicated Ethernet over single mode fiber, not Internet). This is a pretty common broadcast contribution workflow.

Any thoughts? Given a standard HD-SDI implementation AFAIK the smallest unit you work with is an entire frame. I know the custom SDI boards Kierank and the gang over at Open Broadcast Systems have developed allow you to read lines independently so you could get sub-frame latency, but how would that relate to MPEG-2 vs H.264 in this case?

[Emulgator]

Intra-Only says no frame is encoded/decoded related to any other, so can be encoded/decoded independently.
No matter if one or the other encoder can use greater blocksizes, in SDI these blocks are not to be served/transmitted independently so that such part of content could be available sooner or later.
Would still boil down to 1 frame granularity as you said.
Realtime encoding has to be faster than one frame's duration, no matter if content is field-based or not.
So I guess we add the ingestion time for that frame's data,
plus the ms that it takes the encoder to encode that frame,
plus transmitting times from end to end,
plus decoding time, and that gould give MPEG-2 a tiny 1..5ms advantage, because less complex on the same FGPA.