The diversity-ecosystem function relationship is an important topic in ecology but has not received much attention in Arctic environments, and has rarely been tested for its stability in time. We studied the temporal variability of benthic ecosystem functioning at hotspots (sites with high benthic boundary fluxes) and coldspots (sites with lower fluxes) across two years in the Canadian Arctic. Benthic remineralisation function was measured as fluxes of oxygen, silicic acid, phosphate, nitrate and nitrite at the sediment-water interface. In addition we determined sediment pigment concentration and taxonomic and functional macrobenthic diversity. To separate temporal from spatial variability, we sampled the same nine sites from the Mackenzie Shelf to Baffin Bay during the same season (summer or fall) in 2008 and 2009. We observed that temporal variability of benthic remineralisation function at hotspots is higher than at coldspots and that taxonomic and functional macrobenthic diversity did not change significantly between years. Temporal variability of food availability (i.e., sediment surface pigment concentration) seemed higher at coldspot than at hotspot areas. Sediment chlorophyll a (Chl a) concentration, taxonomic richness, total abundance, water depth and abundance of the largest gallery-burrowing polychaete Lumbrineristetraura together explained 42% of the total variation in fluxes. Food supply proxies (i.e., sediment Chl a and depth) split hot- from coldspot stations and explained variation on the axis of temporal variability, and macrofaunal community parameters explained variation mostly along the axis separating eastern from western sites with hot- or coldspot regimes. We conclude that variability in benthic remineralisation function, food supply and diversity will react to climate change on different time scales, and that their interactive effects may hide the detection of progressive change, particularly at hotspots. Time-series of benthic functions and its related parameters should be conducted at both hot- and coldspots to produce reliable predictive models.