Cirrus control the longwave radiative budget of the tropics. For the first time, we quantify the variability in cirrus properties and longwave cloud radiative effects (CREs) that arises from using different bulk ice microphysical parameterizations within a single global storm-resolving model. We run five-day meteorologically-nudged simulations with four commonly-used microphysics schemes (M2005, Thompson, P3 and SAM1MOM) and evaluate them with satellite products and in situ observations. Tropical average longwave CRE varies over 20 W m$^{-2}$ between schemes. Within the Thompson scheme, rapid autoconversion of cloud ice to snow leads to deficient anvil cirrus even with radiatively active snow. SAM1MOM, which uses saturation adjustment for cloud ice, also has deficient anvil cirrus. M2005 and P3 simulate cirrus with realistic frozen water path, and P3 best reproduces observed longwave CRE. Even in those schemes, ice crystal number concentrations commonly hit limiters and lack the observed variability and dependence on frozen water content.