diff --git a/Project.toml b/Project.toml
index a846f4a..7febbda 100644
--- a/Project.toml
+++ b/Project.toml
@@ -8,6 +8,7 @@ AbstractFFTs = "621f4979-c628-5d54-868e-fcf4e3e8185c"
 DocStringExtensions = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 MuladdMacro = "46d2c3a1-f734-5fdb-9937-b9b9aeba4221"
+Polyester = "f517fe37-dbe3-4b94-8317-1923a5111588"
 PrecompileTools = "aea7be01-6a6a-4083-8856-8a6e6704d82a"
 Primes = "27ebfcd6-29c5-5fa9-bf4b-fb8fc14df3ae"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
@@ -21,6 +22,7 @@ ExplicitImports = "1.12"
 FFTW = "1.8"
 LinearAlgebra = "<0.0.1, 1"
 MuladdMacro = "0.2"
+Polyester = "0.7"
 PrecompileTools = "1"
 Primes = "0.5"
 Random = "<0.0.1, 1"
diff --git a/src/FFTA.jl b/src/FFTA.jl
index 3309f17..87ce278 100644
--- a/src/FFTA.jl
+++ b/src/FFTA.jl
@@ -4,6 +4,7 @@ using AbstractFFTs: AbstractFFTs
 using DocStringExtensions: TYPEDEF, TYPEDSIGNATURES
 using LinearAlgebra: LinearAlgebra
 using MuladdMacro: @muladd
+using Polyester: @batch
 using Primes: Primes
 using Reexport: @reexport
 
diff --git a/src/plan.jl b/src/plan.jl
index e519246..85dc76b 100644
--- a/src/plan.jl
+++ b/src/plan.jl
@@ -282,13 +282,13 @@ function _foreach_pencil(f::F, A::AbstractArray{<:Any,N}, ::Val{dim}, workers::V
     total = npre * length(Rpost)
     nt = min(length(workers), total ÷ 2)
     if nt > 1 && total * size(A, dim) >= THREAD_THRESHOLD
-        Base.@sync for c in 1:nt
+        # one chunk per worker, run on Polyester's static thread pool (no task
+        # allocation, the calling thread takes a chunk itself)
+        @batch for c in 1:nt
             lo = (c - 1) * total ÷ nt + 1
             hi = c * total ÷ nt
-            # (the task's worker gets its own name: a variable shared with the
-            # serial branch below would be captured by every task)
             wc = workers[c]
-            Threads.@spawn for k in lo:hi
+            for k in lo:hi
                 ipost, ipre = divrem(k - 1, npre)
                 f(wc, Rpre[ipre + 1], Rpost[ipost + 1])
             end