I am trying to reproduce this with veth to no avail.
I compiled a net-next kernel with Toke's configuration, and simplified
veth to only a client and a server. 

The client upstreams to the server and has cake setup; testing with
tcp_4up on the client. The setup is attached, untar, run vsetup.sh
followed by mm2.sh.



cake with unlimited bandwidth, rtt 100ms:
tc qdisc add dev client.l root cake
                           avg       median          # data pts
Ping (ms) ICMP   :         0.01         0.01 ms              343
TCP upload avg   :     17594.81     17933.86 Mbits/s         301
TCP upload sum   :     70379.25     71735.44 Mbits/s         301
TCP upload::1    :     16148.03     17884.60 Mbits/s         300
TCP upload::2    :     18783.95     20993.83 Mbits/s         300
TCP upload::3    :     17774.41     19560.33 Mbits/s         300
TCP upload::4    :     17672.86     21154.05 Mbits/s         300



cake with bandwidth at 60gbit, rtt 100ms:
tc qdisc add dev client.l root cake bandwidth 60gbit
                           avg       median          # data pts
 Ping (ms) ICMP  :         0.01         0.01 ms              343
 TCP upload avg  :     15971.90     15149.72 Mbits/s         301
 TCP upload sum  :     63887.59     60586.67 Mbits/s         301
 TCP upload::1   :     16903.57     17989.00 Mbits/s         300
 TCP upload::2   :     14248.18     11982.24 Mbits/s         300
 TCP upload::3   :     17196.63     18378.08 Mbits/s         300
 TCP upload::4   :     15539.21     16278.00 Mbits/s         300


George