...

I wasn't in Adelaide, but in earlier meetings, I had the impression that there were at least some implementations using IPv4 as a workaround both to lower layer specific problems (e.g., AAL SNAP/MUX) as well as the hard MTU size limit due to lack of fragmentation capability at the data link layer.

The specific problem that certain people were trying to address was the need to run a multiprotocol encapsulation on top of ATM, thereby (further) lowering the efficiency of ATM while still running IS-IS in their networks.

I don't believe there were any meetings of the ISIS WG in Adelaide.

Correct.

I'm not aware of any implementations of ISIS over IPv4, much less in production networks.

Neither am I. However, there is at least one obvious person that you should consult directly.

As for the MTU issue, most use maximum LSP size of 1492 bytes and even stop Hello padding after initialization.

There is no hard MTU issue. The problem that the spec both created and tried to address was that in running IS-IS over IP, it would be necessary to perform IP fragmentation to deal with IS-IS packets that would otherwise fit in the CLNP MTU. Tony