In an environment where memberOf is enabled on a master and disabled on another master the following happens.

1) Several pieces of data, including invalid data is added to the master with memberOf disabled.
2) GOOD BAD GOOD data is sent to the master with memberOf enabled via replication with the other master.
3) The master with memberOf disabled skips and notes that it skips the BAD data. The master that is sending the replicated data also notes that it skipped the BAD data. Replication is still shown as good.

So:
A) The consumer isn't severing the connection.
B) The supplier should consider schema validation replication error as a hard error and not skip it.

More info and a recreation case for this can be found in email "Re: CASE 01751197". Ludwig was able to reproduce and left these notes

no, I have on the consumer side:

[09/Dec/2016:16:44:34.222683287 +0100] conn=4 op=0 BIND dn="cn=replrepl,cn=config" method=128 version=3
[09/Dec/2016:16:44:34.222905979 +0100] conn=4 op=0 RESULT err=0 tag=97 nentries=0 etime=0 dn="cn=replrepl,cn=config"
[09/Dec/2016:16:44:34.228996479 +0100] conn=4 op=1 SRCH base="" scope=0 filter="(objectClass=)" attrs="supportedControl supportedExtension"
[09/Dec/2016:16:44:34.229495450 +0100] conn=4 op=1 RESULT err=0 tag=101 nentries=1 etime=0
[09/Dec/2016:16:44:34.229661093 +0100] conn=4 op=2 SRCH base="" scope=0 filter="(objectClass=)" attrs="supportedControl supportedExtension"
[09/Dec/2016:16:44:34.230064532 +0100] conn=4 op=2 RESULT err=0 tag=101 nentries=1 etime=0
[09/Dec/2016:16:44:34.230289566 +0100] conn=4 op=3 EXT oid="2.16.840.1.113730.3.5.12" name="replication-multimaster-extop"
[09/Dec/2016:16:44:34.279233418 +0100] conn=4 op=3 RESULT err=0 tag=120 nentries=0 etime=0
[09/Dec/2016:16:44:34.437387789 +0100] conn=4 op=4 ADD dn="cn=g1,dc=example,dc=com"
[09/Dec/2016:16:44:34.555185922 +0100] conn=4 op=4 RESULT err=65 tag=105 nentries=0 etime=0 csn=584ad162000000010000
[09/Dec/2016:16:44:34.654083075 +0100] conn=4 op=5 EXT oid="2.16.840.1.113730.3.5.5" name="replication-multimaster-extop"
[09/Dec/2016:16:44:34.674259915 +0100] conn=4 op=5 RESULT err=0 tag=120 nentries=0 etime=0
[09/Dec/2016:16:44:48.012642473 +0100] conn=4 op=6 EXT oid="2.16.840.1.113730.3.5.12" name="replication-multimaster-extop"
[09/Dec/2016:16:44:48.064483314 +0100] conn=4 op=6 RESULT err=0 tag=120 nentries=0 etime=0
[09/Dec/2016:16:44:48.229068063 +0100] conn=4 op=7 ADD dn="cn=g1,dc=example,dc=com"
[09/Dec/2016:16:44:48.360800339 +0100] conn=4 op=7 RESULT err=65 tag=105 nentries=0 etime=0 csn=584ad162000000010000
[09/Dec/2016:16:44:48.372697051 +0100] conn=4 op=8 ADD dn="cn=yyy,dc=example,dc=com"
[09/Dec/2016:16:44:48.513254149 +0100] conn=4 op=8 RESULT err=0 tag=105 nentries=0 etime=0 csn=584ad170000000010000
[09/Dec/2016:16:44:48.651607165 +0100] conn=4 op=9 EXT oid="2.16.840.1.113730.3.5.5" name="replication-multimaster-extop"
[09/Dec/2016:16:44:48.669904404 +0100] conn=4 op=9 RESULT err=0 tag=120 nentries=0 etime=0
[09/Dec/2016:16:45:48.735018262 +0100] conn=4 op=11 UNBIND
[09/Dec/2016:16:45:48.735053744 +0100] conn=4 op=11 fd=64 closed - U1

and on the supplier:
[09/Dec/2016:16:44:48.468835078 +0100] - DEBUG - NSMMReplicationPlugin - repl5_inc_update_from_op_result - agmt="cn=meTo_localhost.localdomain:38942" (localhost:38942): Consumer failed to replay change (uniqueid 5118e701-be2611e6-88d1a1a4-9fa8583b, CSN 584ad162000000010000): Object class violation (65). Skipping.