diff --git a/builtin/settingtypes.txt b/builtin/settingtypes.txt
index 37cd7d5e2..47cbe38f1 100644
--- a/builtin/settingtypes.txt
+++ b/builtin/settingtypes.txt
@@ -2009,11 +2009,11 @@ transparency_sorting_group_by_buffers (Transparency Sorting Group by Buffers) bo
 cloud_radius (Cloud radius) int 12 8 62
 
 #    Delay between mesh updates on the client in ms. Increasing this will slow
-#    down the rate of mesh updates, thus reducing jitter on slower clients.
-mesh_generation_interval (Mapblock mesh generation delay) int 0 0 50
+#    down the rate of mesh updates, which can help reduce jitter.
+mesh_generation_interval (Mapblock mesh generation delay) int 0 0 25
 
 #    Number of threads to use for mesh generation.
-#    Value of 0 (default) will let Luanti autodetect the number of available threads.
+#    Value of 0 (default) will let Luanti automatically choose the number of threads.
 mesh_generation_threads (Mapblock mesh generation threads) int 0 0 8
 
 #    All mesh buffers with less than this number of vertices will be merged
diff --git a/src/client/mesh_generator_thread.cpp b/src/client/mesh_generator_thread.cpp
index 0a131ba11..5154e8579 100644
--- a/src/client/mesh_generator_thread.cpp
+++ b/src/client/mesh_generator_thread.cpp
@@ -221,18 +221,13 @@ MeshUpdateWorkerThread::MeshUpdateWorkerThread(Client *client, MeshUpdateQueue *
 		UpdateThread("Mesh"), m_client(client), m_queue_in(queue_in), m_manager(manager)
 {
 	m_generation_interval = g_settings->getU16("mesh_generation_interval");
-	m_generation_interval = rangelim(m_generation_interval, 0, 50);
+	m_generation_interval = rangelim(m_generation_interval, 0, 25);
 }
 
 void MeshUpdateWorkerThread::doUpdate()
 {
 	QueuedMeshUpdate *q;
 	while ((q = m_queue_in->pop())) {
-		if (m_generation_interval)
-			sleep_ms(m_generation_interval);
-
-		porting::TriggerMemoryTrim();
-
 		ScopeProfiler sp(g_profiler, "Client: Mesh making (sum)");
 
 		// This generates the mesh:
@@ -249,6 +244,14 @@ void MeshUpdateWorkerThread::doUpdate()
 		m_manager->putResult(r);
 		m_queue_in->done(q->p);
 		delete q;
+		sp.stop();
+
+		porting::TriggerMemoryTrim();
+
+		// do this after we're done so the interval is enforced without
+		// adding extra latency.
+		if (m_generation_interval)
+			sleep_ms(m_generation_interval);
 	}
 }
 
@@ -261,12 +264,12 @@ MeshUpdateManager::MeshUpdateManager(Client *client):
 {
 	int number_of_threads = rangelim(g_settings->getS32("mesh_generation_threads"), 0, 8);
 
-	// Automatically use 33% of the system cores for mesh generation, max 4
+	// Automatically use 25% of the system cores for mesh generation, max 3
 	if (number_of_threads == 0)
-		number_of_threads = MYMIN(4, Thread::getNumberOfProcessors() / 3);
+		number_of_threads = std::min(3U, Thread::getNumberOfProcessors() / 4);
 
 	// use at least one thread
-	number_of_threads = MYMAX(1, number_of_threads);
+	number_of_threads = std::max(1, number_of_threads);
 	infostream << "MeshUpdateManager: using " << number_of_threads << " threads" << std::endl;
 
 	for (int i = 0; i < number_of_threads; i++)
diff --git a/src/profiler.cpp b/src/profiler.cpp
index 3516f4ef3..e965f530e 100644
--- a/src/profiler.cpp
+++ b/src/profiler.cpp
@@ -17,7 +17,7 @@ ScopeProfiler::ScopeProfiler(Profiler *profiler, const std::string &name,
 	m_time1 = porting::getTime(prec);
 }
 
-ScopeProfiler::~ScopeProfiler()
+void ScopeProfiler::stop() noexcept
 {
 	if (!m_profiler)
 		return;
@@ -38,6 +38,8 @@ ScopeProfiler::~ScopeProfiler()
 		m_profiler->max(m_name, duration);
 		break;
 	}
+
+	m_profiler = nullptr; // don't stop a second time
 }
 
 Profiler::Profiler()
diff --git a/src/profiler.h b/src/profiler.h
index 1ecd5ad1c..157d66ef7 100644
--- a/src/profiler.h
+++ b/src/profiler.h
@@ -13,6 +13,8 @@
 #include "threading/mutex_auto_lock.h"
 #include "util/timetaker.h"
 #include "util/numeric.h"      // paging()
+/* FIXME: ^ move this to the .cpp file, it's not needed here */
+#include "util/basic_macros.h"
 
 // Global profiler
 class Profiler;
@@ -108,7 +110,12 @@ public:
 	ScopeProfiler(Profiler *profiler, const std::string &name,
 			ScopeProfilerType type = SPT_ADD,
 			TimePrecision precision = PRECISION_MILLI);
-	~ScopeProfiler();
+	inline ~ScopeProfiler() { stop(); }
+
+	// End profiled scope early
+	void stop() noexcept;
+
+	DISABLE_CLASS_COPY(ScopeProfiler)
 
 private:
 	Profiler *m_profiler = nullptr;