adaptive control tuning parameter

Normally, Lyapunov based adaptive control design guarantees the global
boundedness of estimated paramter and all the closed-loop states. But
in the system parameter estimation--the parameter adaptation/update
law, e.g.,

\dot \theta(t)=-sgn(kp)\gamma_1e(t)y(t)

there is always a tuning coefficient, i.e., \gamma, which theoritically
has nothing to do with stability. However, actually, if this tuning
coefficient is chosen to be too large, the closed-loop system will blow
up.

Any body knows why?  Somebody tells me that discrepancy between theory
and simulation is perhaps that the theory is for a true continuous time
system while the simulation only approximates a continuous time system.
But even for discrete-time system, the same problem exist, so I am
confused....