There are a total of four pyrotechnic bolts holding each SRB to the external tank, one at the forward attach point and three at the aft.
http://science.ksc.nasa.gov/shuttle/technology/sts-newsref/srb.html#srb-sep
Each bolt contains two pyrotechnic initiators. The Soyuz has had at least two failures of pyro bolts linking the descent and service modules. But US pyrotechnic separation mechanisms have been extremely reliable, not because the initiators are redundant, although this plays a role, but because their reliability is actually measured by the classical quality control method of continuous sampling and testing of a sufficient random sample of each production lot. This is a scientifically valid procedure and can achieve any desired level of accuracy. The design has, so far as I know, changed little in recent years but if a lot were found to have any defective units then the manufacturing procedure would be revised to ensure a 100% acceptance rate.
This is in striking contrast to figures such as the "reliability" of the Shuttle or Ares, which has no rigorous statistical basis. The reliability of the Ares, for example, is apparently based on the specification given in the program architecture requirements document. For that matter, the original 1 in 100,000 probabiity of failure of the Shuttle also appears to have originated in a specification rather than by actual testing of components.
Obviously the Shuttle was not as safe as predicted. However the loss-of-crew probability of the Shuttle is now given variously as somewhere between 1/77 and 1/129. Although these figures are fairly close to the historical reliability, this has little bearing on its current reliability, since 1) the failure modes which caused the loss of Challenger and Columbia were eliminated by design changes, and 2) the only consistent observation regarding launch vehicle reliability is that failure rates almost always decrease as the number of launches increase. The Launch Systems Architecture Study advocated using the SRB for the Ares on the basis of its perfect record
since Challenger, completely discounting its role in the loss of 51-L itself. Yet the loss of 51-L is apparently used in the present LOC calculations for the Shuttle, as a pime rationale for killing the Shuttle program. Could these estimates be biased by the apparent desire of program management to get rid of the Shuttle? Considering the lack of statistical rigor across a wide spectrum of risk estimates, W. Edwards Deming, the father of quality control, would be aghast.
The Shuttle now has 129 launches, more than any other manned launch system, although of course the unmanned variants of the Soyuz have considerably more. If there is another failure mode that can cause the loss of a crew, it has not shown itself in 129 flights, so the best estimate of its future LOC rate is certainly less than 1/129.
Constellation supporters would likely say that the Orion is immune to staging pyro bolt failures because of the Launch Abort System, which, according to the "fail safe" philosophy, somewhat magically protects the crew against all failures. But consider this: Suppose the crew actually has time to initiate a LAS abort, and one of the pyro bolts holding the Orion to the service module fails to separate?