|AFAIR without going and looking it up (lazy!) the CD4094 has the
|extra flip-flop and also a latch so data on outputs does not change
|while shifting. I use 2 in series as outputs in one design and 2 x
|CD4021 as equivalent inputs.

The 4094 does indeed have the other latch, and may be a better
part to use for future designs.  The 4021, however, does not
appear to have the extra latch and thus may be subject to the
type of problems described.  On the other hand, when all of the
chips in the shifting chain are on the same board and the clock
edges are sharp and the clock traces feed all the shifters sim-
ilarly, there's not likely to be any problem.

The difficulty comes when, e.g., the clock is fed to shifters on
two separate boards and it's slope-limitted to minimize EMI.
No matter how slow the bit rate, the clock edge will trigger both
of the following events:

[1] The data input of the second device will latch the data from the
    first device.

[2] The data output from the first device will stop outputting the
    bit it was outputting (which is what we want the second device
    to latch) and start outputting new data.

Ideally, [2] will happen before [1], but if different Shmidtt trig-
gers have different threshholds, there's no guarantee that that will
be the case.  Adding a latch whose clock polarity is the opposite of
the shift registers' will cause the two events mentioned above to
happen on opposite clock edges; provided the clock edges do not occur
too close together, there will no longer be any race conditions.

I will mention an important caveat here, however: be careful of any
coupling between the data cascade signal going between the shifters
and the clock wire.  I had some trouble recently with that: the clock
would sometimes pick up a little noise from the switching data signal;
since the clock was itself in transition the noise was apparently en-
ough to cause a very brief (20ns or so) glitch on the clock wire which
the shifter quite happily used as an excuse to shift another bit.

When shifters work, they're wonderful.  But sometimes they can be a
pain in the [bleep]...