The Einstein radius plays a central role in lens studies as it characterises
the strength of gravitational lensing. The distribution of Einstein radii near
the upper cutoff should probe the largest mass concentrations in the universe.
Adopting a triaxial halo model, we compute expected distributions of large
Einstein radii. To assess the cosmic variance, we generate a number of all-sky
Monte-Carlo realisations. We find that the expected largest Einstein radius in
the universe is sensitive to the cosmological model: for a source redshift z=1,
they are 42^{+9}_{-7}, 35^{+8}_{-6}, and 54^{+12}_{-7} arcseconds, assuming
best-fit parameters of the WMAP5, WMAP3 and WMAP1 data, respectively. These
values are broadly consistent with current observations given their
incompleteness. For the same source redshift, we expect in all-sky 35 (WMAP5),
15 (WMAP3), and 150 (WMAP1) clusters that have Einstein radii larger than 20".
Whilst the values of the largest Einstein radii are almost unaffected by the
primordial non-Gaussianity currently of interest, the abundance of large lens
clusters should probe non-Gaussianity competitively with CMB, but only if other
cosmological parameters are well-measured. We also find that these "superlens"
clusters constitute a highly biased population. For instance, a substantial
fraction of these superlens clusters have major axes preferentially aligned
with the line-of-sight. As a consequence, the projected mass distributions of
the clusters are rounder by an ellipticity of 0.2 and have 40%-60% larger
concentrations compared with typical clusters with similar redshifts and
masses. We argue that the large concentration measured in A1689 is consistent
with our model prediction at the 1.2\sigma level. (Abridged)Comment: 17 pages, 12 figures, 1 table, accepted for publication in MNRA