diff --git a/Cargo.lock b/Cargo.lock
index 102516ce..43440ea8 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -729,6 +729,7 @@ dependencies = [
  "md5",
  "memmap2",
  "memtrack",
+ "mimalloc",
  "nestify",
  "nix 0.31.3",
  "object",
diff --git a/Cargo.toml b/Cargo.toml
index 4cc37af8..8bcc2ba8 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -60,6 +60,7 @@ linux-perf-data = { git = "https://github.com/mstange/linux-perf-data.git", rev
 ] } # unreleased main as of 2026-03-19
 debugid = "0.8.0"
 memmap2 = "0.9.10"
+mimalloc = "0.1"
 nix = { version = "0.31.3", features = ["fs", "time", "user"] }
 futures = "0.3.32"
 runner-shared = { path = "crates/runner-shared" }
diff --git a/src/main.rs b/src/main.rs
index 518b8d3e..678ac3ca 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -1,3 +1,9 @@
+// The release binary targets musl, whose allocator is slow on small
+// allocations and serializes threads. Decoding memtrack artifacts allocates per
+// event, which made post-processing take minutes on large artifacts.
+#[global_allocator]
+static GLOBAL: mimalloc::MiMalloc = mimalloc::MiMalloc;
+
 use codspeed_runner::{clean_logger, cli};
 use console::style;
 use log::log_enabled;