Huge cost of ExtendMap on time distribution and filtering per attribute on small range

When data distribution is loaded per datetime

--------------------------------------+

--------------------------------------+

--------------------------------------+

select count from fulldata where id=12675;
----------

----------

----------
1 row in set (4.06 sec)

MaxMemPct-0;
NumTempFiles-0;
TempFileSpace-0B;
ApproxPhyI/O-0;
CacheI/O-3005367;
BlocksTouched-3005367;
PartitionBlocksEliminated-0;
MsgBytesIn-88MB;
MsgBytesOut-704KB;

Desc Mode Table TableOID ReferencedColumns PIO LIO PBE Elapsed Rows
BPS PM fulldata 3108 (idProducer) 0 3005367 0 4.008 751344
TAS UM - - - - - - 4.007 1

When data is loaded per id
select count from fulldataOrdered where id=12675;
----------

----------

----------
1 row in set (0.07 sec)

MariaDB [datagenerated]> select calgetstats();

MaxMemPct-0;
NumTempFiles-0;
TempFileSpace-0B;
ApproxPhyI/O-0;
CacheI/O-16384;
BlocksTouched-16384;
PartitionBlocksEliminated-563697;
MsgBytesIn-492KB;
MsgBytesOut-6KB;
Mode-Distributed

Partition elimination does not work in the first distribution

Desc Mode Table TableOID ReferencedColumns PIO LIO PBE Elapsed Rows
BPS PM fulldataordered 3216 (idProducer) 0 16384 582408 0.078 4096
TAS UM - - - - - - 0.076 1

When comparing both results there is a huge diff
MsgBytesIn 88MB and 751344 Rows sent
492KB MsgBytesIn and 4096 Rows sent
It looks that or the condition is not push down or the full block is send to UM and row extract taking place in UM ?