We consider a mechanism for area preserving Hamiltonian systems which leads to the enhanced probability, P (λ, t), to find small values of the finite time Lyapunov exponent, λ. In our investigation of chaotic dynamical systems we go beyond the linearized stability analysis of nearby divergent trajectories and consider folding of the phase space in the course of chaotic evolution. We show that the spectrum of the Lyapunov exponents F (λ) = lim t→∞ t −1 ln P (λ, t) at the origin has a finite value F (0) = −λ and a slope F ′ (0) ≤ 1. This means that all negative moments of the distribution e −mλt are saturated by rare events with λ → 0. Extensive numerical simulations confirm our findings. An exponential divergency of nearby trajectories in phase space is commonly considered as a paradigm of classical chaos [1,2]. The time evolution of the distance between two trajectories is determined by the stability matrix M(t), whose largest eigenvalue grows exponentially like e λt . For finite time t the value of λ depends on initial conditions and it is called a finite time Lyapunov exponent. The probability distribution of finite time Lyapunov exponents P (λ, t) [2,3,4], especially its behavior at small λ, is important for many applications, where the measured quantity is sensitive to the existence of trajectories staying close for anomalously long time. Examples range from the problems of ocean acoustics [5] and branching of 2d electron flow [6] to Loschmidt echo [7] and mesoscopic superconductivity [8].